Estimating cloud condensation nuclei from aerosol optical properties across diverse environments: observations, models, and prediction approaches

Aerosol-cloud interactions (ACI) remain among the largest sources of uncertainty in assessing anthropogenic impacts on climate (IPCC, 2023), largely due to limited understanding of aerosol sources and their evolution into cloud condensation nuclei (CCN). Reducing this uncertainty requires improved characterization of CCN concentrations and their spatiotemporal variability.

Although CCN measurements are increasingly available at ground-based station, long-term and spatially extensive datasets remain scarce. Harmonized CCN datasets such as those by Schmale et al. (2017) and Andrews et al. (2025) provide quality-assured observations across multiple stations and environments.

To extend CCN information beyond direct measurements, several approaches have been proposed to predict CCN concentrations from more routinely measured aerosol properties, such as aerosol optical properties (AOPs). Using the harmonized Andrews et al. (2025) dataset, Zabala et al. (2025) developed two AOP-based approaches: (i) an empirical parameterization and (ii) a Random Forest (RF) method, based on observations from nine stations (blue in Figure 1). Both methods demonstrate significant potential to extend CCN estimates across space and time.

Harmonized CCN observations have also enabled model evaluation studies. Fanourgakis et al. (2019) evaluated 14 general circulation models against CCN observations from nine stations (orange in Figure 1) over 2011–2015, showing systematic underestimation and substantial variability across environments.

Figure 1. Map of the sites considered in this work.

Motivated by the skill of AOP-based CCN prediction methods, this study applies the two approaches proposed by Zabala et al. (2025) to additional stations with available measurements. The predicted CCN values are evaluated against independent harmonized CCN observations from Schmale et al. (2017) and compared with multimodel CCN estimates reported by Fanourgakis et al. (2019), enabling a consistent assessment across diverse environments.

As an example, Figure 2 shows monthly median CCN concentrations (N_CCN) at 0.5% supersaturation (SS) for the SMEAR (SMR, 61°51'N, 24°17'E, 181 m) station in Finland, including observations, AOP-based predictions and CAM5-MAM3 model simulations. The empirical parameterization and the model generally underestimate N_CCN (median relative biases of -30% and -14%), whereas the RF approach overestimates observations (MRB=75%). Both prediction approaches capture the seasonal cycle, with larger amplitude in the RF estimates. This behavior is consistent across all tested SS.

Figure 2. Monthly median N_CCN (SS=0.5%) at the SMR station from observations, AOP-based predictions, and the CAM5–MAM3 model; shaded area shows the interquartile model range.

Overall, this work demonstrates that AOP-based CCN prediction approaches can reliably extend CCN information beyond observational gaps when evaluated across multiple environments and benchmarked against observations and models. These approaches provide a pathway to improve global CCN datasets, support model evaluation, and reduce uncertainties in ACI in climate models.

This work was supported by the US Department of Energy (DE-SC0022886), the University of Granada (UCE-PP2017-02), and the NUCLEUS (PID2021-128757OB-I00) and MIXDUST (PID2024-160280NB-I00) projects funded by MICIU/AEI, EU NextGenerationEU/PRTR, and FEDER. We acknowledge EBAS (NILU) for the observational data.

References

• Andrews et al. (2025). Sci. Data. 12, 937. Dataset.
• Fanourgakis et al. (2019). Atmos. Chem. Phys., 19, 8591–8617.
• IPCC (2023). Cambridge Uni. Press., Cambridge.
• Schmale et al. (2017). Sci. Data. 4, 937. 170003.
• Zabala et al. (2025). EGUsphere [preprint].